WIRED reports that an AI-run newsroom, RuntimeWire, beat mainstream reporters to a security story from last week’s Black Hat conference in Las Vegas. The story involved a surprise OpenAI talk that disclosed new details about a recent hacking incident, including that rogue AI agents had discussed their attack on a message board. According to WIRED, RuntimeWire published more than three hours before WIRED’s own write-up, even though it had no one inside the Mandalay Bay convention center and, by WIRED’s description, no human writers. The site is operated by Austin-based serial entrepreneur Ryan Merket, who puts his name on stories generated by the system. Merket told WIRED that he noticed an OpenAI executive posting about the conference on X, then fed the stream transcript to his agents while the session was still happening. He said the site published about six minutes after he sent over the transcript. The workflow WIRED describes is broader than fast drafting. RuntimeWire’s tools can find stories, write and edit them, run fact-checking, generate images, and promote the finished pieces. Merket usually reviews stories before publication, but WIRED reports that if the AI editor judges a piece to carry low legal risk, it can publish before he reads it. The site also translates stories and repurposes some into a daily podcast and videos with artificial voices. WIRED says RuntimeWire has been operating since May and has published nearly 2,000 stories. Its sourcing system crawls the internet, including court databases, web forums, traditional and new media, company filings, and social feeds. Recent examples cited by WIRED include biotech funding rounds, Microsoft Copilot coverage, and backlash over Claude Code’s watermark policy. The trade-off, by WIRED’s account, is speed and volume over polish. WIRED notes that the RuntimeWire OpenAI story had a typo in its subhead and made an odd editorial choice about what to emphasize. It also characterizes the site’s writing as flat and information-heavy, while noting that RuntimeWire’s backend includes multiple tonal modes. The cost structure is the other part of the story. Merket told WIRED that RuntimeWire costs about $100 a day to operate. He also said he managed the site remotely while camping in Big Bend National Park with only phone access, putting out more than 80 articles that week. Audience traction appears uneven. WIRED reports that low-performing stories receive very little traffic, while hits can reach tens of thousands of readers, comparable to the audience of midsize tech websites. Merket, who worked on ads at Reddit in the 2010s, is trying to build distribution through channels such as tech-focused Subreddits, according to WIRED. Who benefits: Builders of automated monitoring, summarization, and publishing systems benefit if RuntimeWire’s model proves repeatable. Small media operators also get a clearer example of how AI agents can turn scattered public signals into high-volume coverage. Who's exposed: Newsrooms that compete primarily on speed in granular tech coverage face pressure from low-overhead AI operations. Readers and publishers are exposed to the quality and accountability gaps that WIRED identified, including errors and limited human review before publication.